Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types

Ole-Johan Skrede; Manohar Pradhan; Maria Xepapadakis Isaksen; Tarjei Sveinsgjerd Hveem; Ljiljana Vlatkovic; Arild Nesbakken; Kristina Lindemann; Gunnar B. Kristensen; Jenneke Kasius; Alain G. Zeimet; Odd Terje Brustugun; Lill-Tove Rasmussen Busund; Elin H. Richardsen; Erik Skaaheim Haug; Bjørn Brennhovd; Emma Rewcastle; Melinda Lillesand; Vebjørn Kvikstad; Emiel Janssen; David J. Kerr; Knut Liestøl; Fritz Albregtsen; Andreas Kleppe

doi:10.1038/s41698-026-01311-6

Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types

Ole-Johan Skrede^*
, Manohar Pradhan
, Maria Xepapadakis Isaksen
, Tarjei Sveinsgjerd Hveem
, Ljiljana Vlatkovic
, Arild Nesbakken
, Kristina Lindemann
, Gunnar B. Kristensen
, Jenneke Kasius
, Alain G. Zeimet
, Odd Terje Brustugun
, Lill-Tove Rasmussen Busund
, Elin H. Richardsen
, Erik Skaaheim Haug
, Bjørn Brennhovd
, Emma Rewcastle
, Melinda Lillesand
, Vebjørn Kvikstad
, Emiel Janssen
, David J. Kerr

Knut Liestøl, Fritz Albregtsen, Andreas Kleppe

^*Corresponding author for this work

University of Oslo
University of Amsterdam
Innsbruck Medical University
Vestre Viken Health Trust
University of Tromsø – The Arctic University of Norway
University Hospital of North Norway
Vestfold Hospital Trust
Stavanger University Hospital
University of Stavanger
University of Oxford

Research output: Contribution to journal › Article › Academic › peer-review

4 Downloads (Pure)

Abstract

Deep learning is expected to aid pathologists in tasks such as tumour segmentation. We developed a general tumour segmentation model for histopathological images and examined its performance in different cancer types. The model was developed using over 20,000 whole-slide images from over 4000 patients with colorectal, endometrial, lung, or prostate carcinoma. Performance was validated in pre-planned analyses on external cohorts with over 3000 patients across six cancer types. Exploratory analyses included over 1500 additional patients from The Cancer Genome Atlas. Average Dice coefficient was over 80% in all validation cohorts with en bloc resection specimens and in The Cancer Genome Atlas cohorts. No performance loss was observed when comparing the general model with single-cancer models specialised in cancer types from the development set. In conclusion, extensive and rigorous evaluations demonstrate that generic tumour segmentation by a single model is possible across cancer types, patient populations, sample preparations and slide scanners.

Original language	English
Article number	107
Journal	npj Precision Oncology
Volume	10
Issue number	1
DOIs	https://doi.org/10.1038/s41698-026-01311-6
Publication status	Published - 1 Dec 2026

Access to Document

10.1038/s41698-026-01311-6

Generalisation-of-automatic-tumour-segmentation-in-histopathological-whole-slide-images-across-multiple-cancer-types.pdfFinal published version, 9.31 MBLicence: CC BY

Cite this

Skrede, O.-J., Pradhan, M., Isaksen, M. X., Hveem, T. S., Vlatkovic, L., Nesbakken, A., Lindemann, K., Kristensen, G. B., Kasius, J., Zeimet, A. G., Brustugun, O. T., Busund, L.-T. R., Richardsen, E. H., Haug, E. S., Brennhovd, B., Rewcastle, E., Lillesand, M., Kvikstad, V., Janssen, E., ... Kleppe, A. (2026). Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types. npj Precision Oncology, 10(1), Article 107. https://doi.org/10.1038/s41698-026-01311-6

@article{4ec571f7ac374189a5c39abb6d5cf65f,

title = "Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types",

abstract = "Deep learning is expected to aid pathologists in tasks such as tumour segmentation. We developed a general tumour segmentation model for histopathological images and examined its performance in different cancer types. The model was developed using over 20,000 whole-slide images from over 4000 patients with colorectal, endometrial, lung, or prostate carcinoma. Performance was validated in pre-planned analyses on external cohorts with over 3000 patients across six cancer types. Exploratory analyses included over 1500 additional patients from The Cancer Genome Atlas. Average Dice coefficient was over 80\% in all validation cohorts with en bloc resection specimens and in The Cancer Genome Atlas cohorts. No performance loss was observed when comparing the general model with single-cancer models specialised in cancer types from the development set. In conclusion, extensive and rigorous evaluations demonstrate that generic tumour segmentation by a single model is possible across cancer types, patient populations, sample preparations and slide scanners.",

author = "Ole-Johan Skrede and Manohar Pradhan and Isaksen, \{Maria Xepapadakis\} and Hveem, \{Tarjei Sveinsgjerd\} and Ljiljana Vlatkovic and Arild Nesbakken and Kristina Lindemann and Kristensen, \{Gunnar B.\} and Jenneke Kasius and Zeimet, \{Alain G.\} and Brustugun, \{Odd Terje\} and Busund, \{Lill-Tove Rasmussen\} and Richardsen, \{Elin H.\} and Haug, \{Erik Skaaheim\} and Bj{\o}rn Brennhovd and Emma Rewcastle and Melinda Lillesand and Vebj{\o}rn Kvikstad and Emiel Janssen and Kerr, \{David J.\} and Knut Liest{\o}l and Fritz Albregtsen and Andreas Kleppe",

note = "Publisher Copyright: {\textcopyright} The Author(s) 2026.",

year = "2026",

month = dec,

day = "1",

doi = "10.1038/s41698-026-01311-6",

language = "English",

volume = "10",

journal = "npj Precision Oncology",

issn = "2397-768X",

publisher = "Springer Nature",

number = "1",

}

Skrede, O-J, Pradhan, M, Isaksen, MX, Hveem, TS, Vlatkovic, L, Nesbakken, A, Lindemann, K, Kristensen, GB, Kasius, J, Zeimet, AG, Brustugun, OT, Busund, L-TR, Richardsen, EH, Haug, ES, Brennhovd, B, Rewcastle, E, Lillesand, M, Kvikstad, V, Janssen, E, Kerr, DJ, Liestøl, K, Albregtsen, F & Kleppe, A 2026, 'Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types', npj Precision Oncology, vol. 10, no. 1, 107. https://doi.org/10.1038/s41698-026-01311-6

TY - JOUR

T1 - Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types

AU - Skrede, Ole-Johan

AU - Pradhan, Manohar

AU - Isaksen, Maria Xepapadakis

AU - Hveem, Tarjei Sveinsgjerd

AU - Vlatkovic, Ljiljana

AU - Nesbakken, Arild

AU - Lindemann, Kristina

AU - Kristensen, Gunnar B.

AU - Kasius, Jenneke

AU - Zeimet, Alain G.

AU - Brustugun, Odd Terje

AU - Busund, Lill-Tove Rasmussen

AU - Richardsen, Elin H.

AU - Haug, Erik Skaaheim

AU - Brennhovd, Bjørn

AU - Rewcastle, Emma

AU - Lillesand, Melinda

AU - Kvikstad, Vebjørn

AU - Janssen, Emiel

AU - Kerr, David J.

AU - Liestøl, Knut

AU - Albregtsen, Fritz

AU - Kleppe, Andreas

PY - 2026/12/1

Y1 - 2026/12/1

N2 - Deep learning is expected to aid pathologists in tasks such as tumour segmentation. We developed a general tumour segmentation model for histopathological images and examined its performance in different cancer types. The model was developed using over 20,000 whole-slide images from over 4000 patients with colorectal, endometrial, lung, or prostate carcinoma. Performance was validated in pre-planned analyses on external cohorts with over 3000 patients across six cancer types. Exploratory analyses included over 1500 additional patients from The Cancer Genome Atlas. Average Dice coefficient was over 80% in all validation cohorts with en bloc resection specimens and in The Cancer Genome Atlas cohorts. No performance loss was observed when comparing the general model with single-cancer models specialised in cancer types from the development set. In conclusion, extensive and rigorous evaluations demonstrate that generic tumour segmentation by a single model is possible across cancer types, patient populations, sample preparations and slide scanners.

AB - Deep learning is expected to aid pathologists in tasks such as tumour segmentation. We developed a general tumour segmentation model for histopathological images and examined its performance in different cancer types. The model was developed using over 20,000 whole-slide images from over 4000 patients with colorectal, endometrial, lung, or prostate carcinoma. Performance was validated in pre-planned analyses on external cohorts with over 3000 patients across six cancer types. Exploratory analyses included over 1500 additional patients from The Cancer Genome Atlas. Average Dice coefficient was over 80% in all validation cohorts with en bloc resection specimens and in The Cancer Genome Atlas cohorts. No performance loss was observed when comparing the general model with single-cancer models specialised in cancer types from the development set. In conclusion, extensive and rigorous evaluations demonstrate that generic tumour segmentation by a single model is possible across cancer types, patient populations, sample preparations and slide scanners.

UR - https://www.scopus.com/pages/publications/105033838505

U2 - 10.1038/s41698-026-01311-6

DO - 10.1038/s41698-026-01311-6

M3 - Article

C2 - 41634087

SN - 2397-768X

VL - 10

JO - npj Precision Oncology

JF - npj Precision Oncology

IS - 1

M1 - 107

ER -

Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types

Abstract

Access to Document

Other files and links

Fingerprint

Cite this